xref: /freebsd/sys/fs/ext2fs/ext2_vnops.c (revision 41059135ce931c0f1014a999ffabc6bc470ce856)
1 /*-
2  *  modified for EXT2FS support in Lites 1.1
3  *
4  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
5  *  University of Utah, Department of Computer Science
6  */
7 /*-
8  * Copyright (c) 1982, 1986, 1989, 1993
9  *	The Regents of the University of California.  All rights reserved.
10  * (c) UNIX System Laboratories, Inc.
11  * All or some portions of this file are derived from material licensed
12  * to the University of California by American Telephone and Telegraph
13  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
14  * the permission of UNIX System Laboratories, Inc.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
19  * 1. Redistributions of source code must retain the above copyright
20  *    notice, this list of conditions and the following disclaimer.
21  * 2. Redistributions in binary form must reproduce the above copyright
22  *    notice, this list of conditions and the following disclaimer in the
23  *    documentation and/or other materials provided with the distribution.
24  * 3. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	@(#)ufs_vnops.c	8.7 (Berkeley) 2/3/94
41  *	@(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
42  * $FreeBSD$
43  */
44 
45 #include "opt_suiddir.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/fcntl.h>
51 #include <sys/filio.h>
52 #include <sys/stat.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/endian.h>
56 #include <sys/priv.h>
57 #include <sys/rwlock.h>
58 #include <sys/mount.h>
59 #include <sys/unistd.h>
60 #include <sys/time.h>
61 #include <sys/vnode.h>
62 #include <sys/namei.h>
63 #include <sys/lockf.h>
64 #include <sys/event.h>
65 #include <sys/conf.h>
66 #include <sys/file.h>
67 #include <sys/extattr.h>
68 #include <sys/vmmeter.h>
69 
70 #include <vm/vm.h>
71 #include <vm/vm_param.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_pager.h>
76 #include <vm/vnode_pager.h>
77 
78 #include "opt_directio.h"
79 
80 #include <ufs/ufs/dir.h>
81 
82 #include <fs/ext2fs/fs.h>
83 #include <fs/ext2fs/inode.h>
84 #include <fs/ext2fs/ext2_extern.h>
85 #include <fs/ext2fs/ext2fs.h>
86 #include <fs/ext2fs/ext2_dinode.h>
87 #include <fs/ext2fs/ext2_dir.h>
88 #include <fs/ext2fs/ext2_mount.h>
89 #include <fs/ext2fs/ext2_extattr.h>
90 
91 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
92 static void ext2_itimes_locked(struct vnode *);
93 static int ext4_ext_read(struct vop_read_args *);
94 static int ext2_ind_read(struct vop_read_args *);
95 
96 static vop_access_t	ext2_access;
97 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
98 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
99     struct thread *);
100 static vop_close_t	ext2_close;
101 static vop_create_t	ext2_create;
102 static vop_fsync_t	ext2_fsync;
103 static vop_getattr_t	ext2_getattr;
104 static vop_ioctl_t	ext2_ioctl;
105 static vop_link_t	ext2_link;
106 static vop_mkdir_t	ext2_mkdir;
107 static vop_mknod_t	ext2_mknod;
108 static vop_open_t	ext2_open;
109 static vop_pathconf_t	ext2_pathconf;
110 static vop_print_t	ext2_print;
111 static vop_read_t	ext2_read;
112 static vop_readlink_t	ext2_readlink;
113 static vop_remove_t	ext2_remove;
114 static vop_rename_t	ext2_rename;
115 static vop_rmdir_t	ext2_rmdir;
116 static vop_setattr_t	ext2_setattr;
117 static vop_strategy_t	ext2_strategy;
118 static vop_symlink_t	ext2_symlink;
119 static vop_write_t	ext2_write;
120 static vop_deleteextattr_t	ext2_deleteextattr;
121 static vop_getextattr_t	ext2_getextattr;
122 static vop_listextattr_t	ext2_listextattr;
123 static vop_setextattr_t	ext2_setextattr;
124 static vop_vptofh_t	ext2_vptofh;
125 static vop_close_t	ext2fifo_close;
126 static vop_kqfilter_t	ext2fifo_kqfilter;
127 
128 /* Global vfs data structures for ext2. */
129 struct vop_vector ext2_vnodeops = {
130 	.vop_default =		&default_vnodeops,
131 	.vop_access =		ext2_access,
132 	.vop_bmap =		ext2_bmap,
133 	.vop_cachedlookup =	ext2_lookup,
134 	.vop_close =		ext2_close,
135 	.vop_create =		ext2_create,
136 	.vop_fsync =		ext2_fsync,
137 	.vop_getpages =		vnode_pager_local_getpages,
138 	.vop_getpages_async =	vnode_pager_local_getpages_async,
139 	.vop_getattr =		ext2_getattr,
140 	.vop_inactive =		ext2_inactive,
141 	.vop_ioctl =		ext2_ioctl,
142 	.vop_link =		ext2_link,
143 	.vop_lookup =		vfs_cache_lookup,
144 	.vop_mkdir =		ext2_mkdir,
145 	.vop_mknod =		ext2_mknod,
146 	.vop_open =		ext2_open,
147 	.vop_pathconf =		ext2_pathconf,
148 	.vop_poll =		vop_stdpoll,
149 	.vop_print =		ext2_print,
150 	.vop_read =		ext2_read,
151 	.vop_readdir =		ext2_readdir,
152 	.vop_readlink =		ext2_readlink,
153 	.vop_reallocblks =	ext2_reallocblks,
154 	.vop_reclaim =		ext2_reclaim,
155 	.vop_remove =		ext2_remove,
156 	.vop_rename =		ext2_rename,
157 	.vop_rmdir =		ext2_rmdir,
158 	.vop_setattr =		ext2_setattr,
159 	.vop_strategy =		ext2_strategy,
160 	.vop_symlink =		ext2_symlink,
161 	.vop_write =		ext2_write,
162 	.vop_deleteextattr =	ext2_deleteextattr,
163 	.vop_getextattr =	ext2_getextattr,
164 	.vop_listextattr =	ext2_listextattr,
165 	.vop_setextattr =	ext2_setextattr,
166 	.vop_vptofh =		ext2_vptofh,
167 };
168 
169 struct vop_vector ext2_fifoops = {
170 	.vop_default =		&fifo_specops,
171 	.vop_access =		ext2_access,
172 	.vop_close =		ext2fifo_close,
173 	.vop_fsync =		ext2_fsync,
174 	.vop_getattr =		ext2_getattr,
175 	.vop_inactive =		ext2_inactive,
176 	.vop_kqfilter =		ext2fifo_kqfilter,
177 	.vop_print =		ext2_print,
178 	.vop_read =		VOP_PANIC,
179 	.vop_reclaim =		ext2_reclaim,
180 	.vop_setattr =		ext2_setattr,
181 	.vop_write =		VOP_PANIC,
182 	.vop_vptofh =		ext2_vptofh,
183 };
184 
185 /*
186  * A virgin directory (no blushing please).
187  * Note that the type and namlen fields are reversed relative to ext2.
188  * Also, we don't use `struct odirtemplate', since it would just cause
189  * endianness problems.
190  */
191 static struct dirtemplate mastertemplate = {
192 	0, 12, 1, EXT2_FT_DIR, ".",
193 	0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
194 };
195 static struct dirtemplate omastertemplate = {
196 	0, 12, 1, EXT2_FT_UNKNOWN, ".",
197 	0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
198 };
199 
200 static void
201 ext2_itimes_locked(struct vnode *vp)
202 {
203 	struct inode *ip;
204 	struct timespec ts;
205 
206 	ASSERT_VI_LOCKED(vp, __func__);
207 
208 	ip = VTOI(vp);
209 	if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
210 		return;
211 	if ((vp->v_type == VBLK || vp->v_type == VCHR))
212 		ip->i_flag |= IN_LAZYMOD;
213 	else
214 		ip->i_flag |= IN_MODIFIED;
215 	if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
216 		vfs_timestamp(&ts);
217 		if (ip->i_flag & IN_ACCESS) {
218 			ip->i_atime = ts.tv_sec;
219 			ip->i_atimensec = ts.tv_nsec;
220 		}
221 		if (ip->i_flag & IN_UPDATE) {
222 			ip->i_mtime = ts.tv_sec;
223 			ip->i_mtimensec = ts.tv_nsec;
224 			ip->i_modrev++;
225 		}
226 		if (ip->i_flag & IN_CHANGE) {
227 			ip->i_ctime = ts.tv_sec;
228 			ip->i_ctimensec = ts.tv_nsec;
229 		}
230 	}
231 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
232 }
233 
234 void
235 ext2_itimes(struct vnode *vp)
236 {
237 
238 	VI_LOCK(vp);
239 	ext2_itimes_locked(vp);
240 	VI_UNLOCK(vp);
241 }
242 
243 /*
244  * Create a regular file
245  */
246 static int
247 ext2_create(struct vop_create_args *ap)
248 {
249 	int error;
250 
251 	error =
252 	    ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
253 	    ap->a_dvp, ap->a_vpp, ap->a_cnp);
254 	if (error != 0)
255 		return (error);
256 	if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
257 		cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
258 	return (0);
259 }
260 
261 static int
262 ext2_open(struct vop_open_args *ap)
263 {
264 
265 	if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
266 		return (EOPNOTSUPP);
267 
268 	/*
269 	 * Files marked append-only must be opened for appending.
270 	 */
271 	if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
272 	    (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
273 		return (EPERM);
274 
275 	vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
276 
277 	return (0);
278 }
279 
280 /*
281  * Close called.
282  *
283  * Update the times on the inode.
284  */
285 static int
286 ext2_close(struct vop_close_args *ap)
287 {
288 	struct vnode *vp = ap->a_vp;
289 
290 	VI_LOCK(vp);
291 	if (vp->v_usecount > 1)
292 		ext2_itimes_locked(vp);
293 	VI_UNLOCK(vp);
294 	return (0);
295 }
296 
297 static int
298 ext2_access(struct vop_access_args *ap)
299 {
300 	struct vnode *vp = ap->a_vp;
301 	struct inode *ip = VTOI(vp);
302 	accmode_t accmode = ap->a_accmode;
303 	int error;
304 
305 	if (vp->v_type == VBLK || vp->v_type == VCHR)
306 		return (EOPNOTSUPP);
307 
308 	/*
309 	 * Disallow write attempts on read-only file systems;
310 	 * unless the file is a socket, fifo, or a block or
311 	 * character device resident on the file system.
312 	 */
313 	if (accmode & VWRITE) {
314 		switch (vp->v_type) {
315 		case VDIR:
316 		case VLNK:
317 		case VREG:
318 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
319 				return (EROFS);
320 			break;
321 		default:
322 			break;
323 		}
324 	}
325 
326 	/* If immutable bit set, nobody gets to write it. */
327 	if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
328 		return (EPERM);
329 
330 	error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
331 	    ap->a_accmode, ap->a_cred, NULL);
332 	return (error);
333 }
334 
335 static int
336 ext2_getattr(struct vop_getattr_args *ap)
337 {
338 	struct vnode *vp = ap->a_vp;
339 	struct inode *ip = VTOI(vp);
340 	struct vattr *vap = ap->a_vap;
341 
342 	ext2_itimes(vp);
343 	/*
344 	 * Copy from inode table
345 	 */
346 	vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
347 	vap->va_fileid = ip->i_number;
348 	vap->va_mode = ip->i_mode & ~IFMT;
349 	vap->va_nlink = ip->i_nlink;
350 	vap->va_uid = ip->i_uid;
351 	vap->va_gid = ip->i_gid;
352 	vap->va_rdev = ip->i_rdev;
353 	vap->va_size = ip->i_size;
354 	vap->va_atime.tv_sec = ip->i_atime;
355 	vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
356 	vap->va_mtime.tv_sec = ip->i_mtime;
357 	vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
358 	vap->va_ctime.tv_sec = ip->i_ctime;
359 	vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
360 	if E2DI_HAS_XTIME(ip) {
361 		vap->va_birthtime.tv_sec = ip->i_birthtime;
362 		vap->va_birthtime.tv_nsec = ip->i_birthnsec;
363 	}
364 	vap->va_flags = ip->i_flags;
365 	vap->va_gen = ip->i_gen;
366 	vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
367 	vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
368 	vap->va_type = IFTOVT(ip->i_mode);
369 	vap->va_filerev = ip->i_modrev;
370 	return (0);
371 }
372 
373 /*
374  * Set attribute vnode op. called from several syscalls
375  */
376 static int
377 ext2_setattr(struct vop_setattr_args *ap)
378 {
379 	struct vattr *vap = ap->a_vap;
380 	struct vnode *vp = ap->a_vp;
381 	struct inode *ip = VTOI(vp);
382 	struct ucred *cred = ap->a_cred;
383 	struct thread *td = curthread;
384 	int error;
385 
386 	/*
387 	 * Check for unsettable attributes.
388 	 */
389 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
390 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
391 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
392 	    ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
393 		return (EINVAL);
394 	}
395 	if (vap->va_flags != VNOVAL) {
396 		/* Disallow flags not supported by ext2fs. */
397 		if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
398 			return (EOPNOTSUPP);
399 
400 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
401 			return (EROFS);
402 		/*
403 		 * Callers may only modify the file flags on objects they
404 		 * have VADMIN rights for.
405 		 */
406 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
407 			return (error);
408 		/*
409 		 * Unprivileged processes and privileged processes in
410 		 * jail() are not permitted to unset system flags, or
411 		 * modify flags if any system flags are set.
412 		 * Privileged non-jail processes may not modify system flags
413 		 * if securelevel > 0 and any existing system flags are set.
414 		 */
415 		if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
416 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) {
417 				error = securelevel_gt(cred, 0);
418 				if (error)
419 					return (error);
420 			}
421 		} else {
422 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
423 			    ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
424 				return (EPERM);
425 		}
426 		ip->i_flags = vap->va_flags;
427 		ip->i_flag |= IN_CHANGE;
428 		if (ip->i_flags & (IMMUTABLE | APPEND))
429 			return (0);
430 	}
431 	if (ip->i_flags & (IMMUTABLE | APPEND))
432 		return (EPERM);
433 	/*
434 	 * Go through the fields and update iff not VNOVAL.
435 	 */
436 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
437 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
438 			return (EROFS);
439 		if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
440 		    td)) != 0)
441 			return (error);
442 	}
443 	if (vap->va_size != VNOVAL) {
444 		/*
445 		 * Disallow write attempts on read-only file systems;
446 		 * unless the file is a socket, fifo, or a block or
447 		 * character device resident on the file system.
448 		 */
449 		switch (vp->v_type) {
450 		case VDIR:
451 			return (EISDIR);
452 		case VLNK:
453 		case VREG:
454 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
455 				return (EROFS);
456 			break;
457 		default:
458 			break;
459 		}
460 		if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
461 			return (error);
462 	}
463 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
464 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
465 			return (EROFS);
466 		/*
467 		 * From utimes(2):
468 		 * If times is NULL, ... The caller must be the owner of
469 		 * the file, have permission to write the file, or be the
470 		 * super-user.
471 		 * If times is non-NULL, ... The caller must be the owner of
472 		 * the file or be the super-user.
473 		 */
474 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
475 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
476 		    (error = VOP_ACCESS(vp, VWRITE, cred, td))))
477 			return (error);
478 		ip->i_flag |= IN_CHANGE | IN_MODIFIED;
479 		if (vap->va_atime.tv_sec != VNOVAL) {
480 			ip->i_flag &= ~IN_ACCESS;
481 			ip->i_atime = vap->va_atime.tv_sec;
482 			ip->i_atimensec = vap->va_atime.tv_nsec;
483 		}
484 		if (vap->va_mtime.tv_sec != VNOVAL) {
485 			ip->i_flag &= ~IN_UPDATE;
486 			ip->i_mtime = vap->va_mtime.tv_sec;
487 			ip->i_mtimensec = vap->va_mtime.tv_nsec;
488 		}
489 		ip->i_birthtime = vap->va_birthtime.tv_sec;
490 		ip->i_birthnsec = vap->va_birthtime.tv_nsec;
491 		error = ext2_update(vp, 0);
492 		if (error)
493 			return (error);
494 	}
495 	error = 0;
496 	if (vap->va_mode != (mode_t)VNOVAL) {
497 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
498 			return (EROFS);
499 		error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
500 	}
501 	return (error);
502 }
503 
504 /*
505  * Change the mode on a file.
506  * Inode must be locked before calling.
507  */
508 static int
509 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
510 {
511 	struct inode *ip = VTOI(vp);
512 	int error;
513 
514 	/*
515 	 * To modify the permissions on a file, must possess VADMIN
516 	 * for that file.
517 	 */
518 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
519 		return (error);
520 	/*
521 	 * Privileged processes may set the sticky bit on non-directories,
522 	 * as well as set the setgid bit on a file with a group that the
523 	 * process is not a member of.
524 	 */
525 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
526 		error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
527 		if (error)
528 			return (EFTYPE);
529 	}
530 	if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
531 		error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
532 		if (error)
533 			return (error);
534 	}
535 	ip->i_mode &= ~ALLPERMS;
536 	ip->i_mode |= (mode & ALLPERMS);
537 	ip->i_flag |= IN_CHANGE;
538 	return (0);
539 }
540 
541 /*
542  * Perform chown operation on inode ip;
543  * inode must be locked prior to call.
544  */
545 static int
546 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
547     struct thread *td)
548 {
549 	struct inode *ip = VTOI(vp);
550 	uid_t ouid;
551 	gid_t ogid;
552 	int error = 0;
553 
554 	if (uid == (uid_t)VNOVAL)
555 		uid = ip->i_uid;
556 	if (gid == (gid_t)VNOVAL)
557 		gid = ip->i_gid;
558 	/*
559 	 * To modify the ownership of a file, must possess VADMIN
560 	 * for that file.
561 	 */
562 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
563 		return (error);
564 	/*
565 	 * To change the owner of a file, or change the group of a file
566 	 * to a group of which we are not a member, the caller must
567 	 * have privilege.
568 	 */
569 	if (uid != ip->i_uid || (gid != ip->i_gid &&
570 	    !groupmember(gid, cred))) {
571 		error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
572 		if (error)
573 			return (error);
574 	}
575 	ogid = ip->i_gid;
576 	ouid = ip->i_uid;
577 	ip->i_gid = gid;
578 	ip->i_uid = uid;
579 	ip->i_flag |= IN_CHANGE;
580 	if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
581 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0)
582 			ip->i_mode &= ~(ISUID | ISGID);
583 	}
584 	return (0);
585 }
586 
587 /*
588  * Synch an open file.
589  */
590 /* ARGSUSED */
591 static int
592 ext2_fsync(struct vop_fsync_args *ap)
593 {
594 	/*
595 	 * Flush all dirty buffers associated with a vnode.
596 	 */
597 
598 	vop_stdfsync(ap);
599 
600 	return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
601 }
602 
603 /*
604  * Mknod vnode call
605  */
606 /* ARGSUSED */
607 static int
608 ext2_mknod(struct vop_mknod_args *ap)
609 {
610 	struct vattr *vap = ap->a_vap;
611 	struct vnode **vpp = ap->a_vpp;
612 	struct inode *ip;
613 	ino_t ino;
614 	int error;
615 
616 	error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
617 	    ap->a_dvp, vpp, ap->a_cnp);
618 	if (error)
619 		return (error);
620 	ip = VTOI(*vpp);
621 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
622 	if (vap->va_rdev != VNOVAL) {
623 		/*
624 		 * Want to be able to use this to make badblock
625 		 * inodes, so don't truncate the dev number.
626 		 */
627 		ip->i_rdev = vap->va_rdev;
628 	}
629 	/*
630 	 * Remove inode, then reload it through VFS_VGET so it is
631 	 * checked to see if it is an alias of an existing entry in
632 	 * the inode cache.	 XXX I don't believe this is necessary now.
633 	 */
634 	(*vpp)->v_type = VNON;
635 	ino = ip->i_number;	/* Save this before vgone() invalidates ip. */
636 	vgone(*vpp);
637 	vput(*vpp);
638 	error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
639 	if (error) {
640 		*vpp = NULL;
641 		return (error);
642 	}
643 	return (0);
644 }
645 
646 static int
647 ext2_remove(struct vop_remove_args *ap)
648 {
649 	struct inode *ip;
650 	struct vnode *vp = ap->a_vp;
651 	struct vnode *dvp = ap->a_dvp;
652 	int error;
653 
654 	ip = VTOI(vp);
655 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
656 	    (VTOI(dvp)->i_flags & APPEND)) {
657 		error = EPERM;
658 		goto out;
659 	}
660 	error = ext2_dirremove(dvp, ap->a_cnp);
661 	if (error == 0) {
662 		ip->i_nlink--;
663 		ip->i_flag |= IN_CHANGE;
664 	}
665 out:
666 	return (error);
667 }
668 
669 /*
670  * link vnode call
671  */
672 static int
673 ext2_link(struct vop_link_args *ap)
674 {
675 	struct vnode *vp = ap->a_vp;
676 	struct vnode *tdvp = ap->a_tdvp;
677 	struct componentname *cnp = ap->a_cnp;
678 	struct inode *ip;
679 	int error;
680 
681 #ifdef INVARIANTS
682 	if ((cnp->cn_flags & HASBUF) == 0)
683 		panic("ext2_link: no name");
684 #endif
685 	ip = VTOI(vp);
686 	if ((nlink_t)ip->i_nlink >= EXT2_LINK_MAX) {
687 		error = EMLINK;
688 		goto out;
689 	}
690 	if (ip->i_flags & (IMMUTABLE | APPEND)) {
691 		error = EPERM;
692 		goto out;
693 	}
694 	ip->i_nlink++;
695 	ip->i_flag |= IN_CHANGE;
696 	error = ext2_update(vp, !DOINGASYNC(vp));
697 	if (!error)
698 		error = ext2_direnter(ip, tdvp, cnp);
699 	if (error) {
700 		ip->i_nlink--;
701 		ip->i_flag |= IN_CHANGE;
702 	}
703 out:
704 	return (error);
705 }
706 
707 /*
708  * Rename system call.
709  * 	rename("foo", "bar");
710  * is essentially
711  *	unlink("bar");
712  *	link("foo", "bar");
713  *	unlink("foo");
714  * but ``atomically''.  Can't do full commit without saving state in the
715  * inode on disk which isn't feasible at this time.  Best we can do is
716  * always guarantee the target exists.
717  *
718  * Basic algorithm is:
719  *
720  * 1) Bump link count on source while we're linking it to the
721  *    target.  This also ensure the inode won't be deleted out
722  *    from underneath us while we work (it may be truncated by
723  *    a concurrent `trunc' or `open' for creation).
724  * 2) Link source to destination.  If destination already exists,
725  *    delete it first.
726  * 3) Unlink source reference to inode if still around. If a
727  *    directory was moved and the parent of the destination
728  *    is different from the source, patch the ".." entry in the
729  *    directory.
730  */
731 static int
732 ext2_rename(struct vop_rename_args *ap)
733 {
734 	struct vnode *tvp = ap->a_tvp;
735 	struct vnode *tdvp = ap->a_tdvp;
736 	struct vnode *fvp = ap->a_fvp;
737 	struct vnode *fdvp = ap->a_fdvp;
738 	struct componentname *tcnp = ap->a_tcnp;
739 	struct componentname *fcnp = ap->a_fcnp;
740 	struct inode *ip, *xp, *dp;
741 	struct dirtemplate dirbuf;
742 	int doingdirectory = 0, oldparent = 0, newparent = 0;
743 	int error = 0;
744 	u_char namlen;
745 
746 #ifdef INVARIANTS
747 	if ((tcnp->cn_flags & HASBUF) == 0 ||
748 	    (fcnp->cn_flags & HASBUF) == 0)
749 		panic("ext2_rename: no name");
750 #endif
751 	/*
752 	 * Check for cross-device rename.
753 	 */
754 	if ((fvp->v_mount != tdvp->v_mount) ||
755 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
756 		error = EXDEV;
757 abortit:
758 		if (tdvp == tvp)
759 			vrele(tdvp);
760 		else
761 			vput(tdvp);
762 		if (tvp)
763 			vput(tvp);
764 		vrele(fdvp);
765 		vrele(fvp);
766 		return (error);
767 	}
768 
769 	if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
770 	    (VTOI(tdvp)->i_flags & APPEND))) {
771 		error = EPERM;
772 		goto abortit;
773 	}
774 
775 	/*
776 	 * Renaming a file to itself has no effect.  The upper layers should
777 	 * not call us in that case.  Temporarily just warn if they do.
778 	 */
779 	if (fvp == tvp) {
780 		printf("ext2_rename: fvp == tvp (can't happen)\n");
781 		error = 0;
782 		goto abortit;
783 	}
784 
785 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
786 		goto abortit;
787 	dp = VTOI(fdvp);
788 	ip = VTOI(fvp);
789 	if (ip->i_nlink >= EXT2_LINK_MAX) {
790 		VOP_UNLOCK(fvp, 0);
791 		error = EMLINK;
792 		goto abortit;
793 	}
794 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
795 	    || (dp->i_flags & APPEND)) {
796 		VOP_UNLOCK(fvp, 0);
797 		error = EPERM;
798 		goto abortit;
799 	}
800 	if ((ip->i_mode & IFMT) == IFDIR) {
801 		/*
802 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
803 		 */
804 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
805 		    dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
806 		    (ip->i_flag & IN_RENAME)) {
807 			VOP_UNLOCK(fvp, 0);
808 			error = EINVAL;
809 			goto abortit;
810 		}
811 		ip->i_flag |= IN_RENAME;
812 		oldparent = dp->i_number;
813 		doingdirectory++;
814 	}
815 	vrele(fdvp);
816 
817 	/*
818 	 * When the target exists, both the directory
819 	 * and target vnodes are returned locked.
820 	 */
821 	dp = VTOI(tdvp);
822 	xp = NULL;
823 	if (tvp)
824 		xp = VTOI(tvp);
825 
826 	/*
827 	 * 1) Bump link count while we're moving stuff
828 	 *    around.  If we crash somewhere before
829 	 *    completing our work, the link count
830 	 *    may be wrong, but correctable.
831 	 */
832 	ip->i_nlink++;
833 	ip->i_flag |= IN_CHANGE;
834 	if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
835 		VOP_UNLOCK(fvp, 0);
836 		goto bad;
837 	}
838 
839 	/*
840 	 * If ".." must be changed (ie the directory gets a new
841 	 * parent) then the source directory must not be in the
842 	 * directory hierarchy above the target, as this would
843 	 * orphan everything below the source directory. Also
844 	 * the user must have write permission in the source so
845 	 * as to be able to change "..". We must repeat the call
846 	 * to namei, as the parent directory is unlocked by the
847 	 * call to checkpath().
848 	 */
849 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
850 	VOP_UNLOCK(fvp, 0);
851 	if (oldparent != dp->i_number)
852 		newparent = dp->i_number;
853 	if (doingdirectory && newparent) {
854 		if (error)	/* write access check above */
855 			goto bad;
856 		if (xp != NULL)
857 			vput(tvp);
858 		error = ext2_checkpath(ip, dp, tcnp->cn_cred);
859 		if (error)
860 			goto out;
861 		VREF(tdvp);
862 		error = relookup(tdvp, &tvp, tcnp);
863 		if (error)
864 			goto out;
865 		vrele(tdvp);
866 		dp = VTOI(tdvp);
867 		xp = NULL;
868 		if (tvp)
869 			xp = VTOI(tvp);
870 	}
871 	/*
872 	 * 2) If target doesn't exist, link the target
873 	 *    to the source and unlink the source.
874 	 *    Otherwise, rewrite the target directory
875 	 *    entry to reference the source inode and
876 	 *    expunge the original entry's existence.
877 	 */
878 	if (xp == NULL) {
879 		if (dp->i_devvp != ip->i_devvp)
880 			panic("ext2_rename: EXDEV");
881 		/*
882 		 * Account for ".." in new directory.
883 		 * When source and destination have the same
884 		 * parent we don't fool with the link count.
885 		 */
886 		if (doingdirectory && newparent) {
887 			if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
888 				error = EMLINK;
889 				goto bad;
890 			}
891 			dp->i_nlink++;
892 			dp->i_flag |= IN_CHANGE;
893 			error = ext2_update(tdvp, !DOINGASYNC(tdvp));
894 			if (error)
895 				goto bad;
896 		}
897 		error = ext2_direnter(ip, tdvp, tcnp);
898 		if (error) {
899 			if (doingdirectory && newparent) {
900 				dp->i_nlink--;
901 				dp->i_flag |= IN_CHANGE;
902 				(void)ext2_update(tdvp, 1);
903 			}
904 			goto bad;
905 		}
906 		vput(tdvp);
907 	} else {
908 		if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
909 			panic("ext2_rename: EXDEV");
910 		/*
911 		 * Short circuit rename(foo, foo).
912 		 */
913 		if (xp->i_number == ip->i_number)
914 			panic("ext2_rename: same file");
915 		/*
916 		 * If the parent directory is "sticky", then the user must
917 		 * own the parent directory, or the destination of the rename,
918 		 * otherwise the destination may not be changed (except by
919 		 * root). This implements append-only directories.
920 		 */
921 		if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
922 		    tcnp->cn_cred->cr_uid != dp->i_uid &&
923 		    xp->i_uid != tcnp->cn_cred->cr_uid) {
924 			error = EPERM;
925 			goto bad;
926 		}
927 		/*
928 		 * Target must be empty if a directory and have no links
929 		 * to it. Also, ensure source and target are compatible
930 		 * (both directories, or both not directories).
931 		 */
932 		if ((xp->i_mode & IFMT) == IFDIR) {
933 			if (!ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
934 			    xp->i_nlink > 2) {
935 				error = ENOTEMPTY;
936 				goto bad;
937 			}
938 			if (!doingdirectory) {
939 				error = ENOTDIR;
940 				goto bad;
941 			}
942 			cache_purge(tdvp);
943 		} else if (doingdirectory) {
944 			error = EISDIR;
945 			goto bad;
946 		}
947 		error = ext2_dirrewrite(dp, ip, tcnp);
948 		if (error)
949 			goto bad;
950 		/*
951 		 * If the target directory is in the same
952 		 * directory as the source directory,
953 		 * decrement the link count on the parent
954 		 * of the target directory.
955 		 */
956 		if (doingdirectory && !newparent) {
957 			dp->i_nlink--;
958 			dp->i_flag |= IN_CHANGE;
959 		}
960 		vput(tdvp);
961 		/*
962 		 * Adjust the link count of the target to
963 		 * reflect the dirrewrite above.  If this is
964 		 * a directory it is empty and there are
965 		 * no links to it, so we can squash the inode and
966 		 * any space associated with it.  We disallowed
967 		 * renaming over top of a directory with links to
968 		 * it above, as the remaining link would point to
969 		 * a directory without "." or ".." entries.
970 		 */
971 		xp->i_nlink--;
972 		if (doingdirectory) {
973 			if (--xp->i_nlink != 0)
974 				panic("ext2_rename: linked directory");
975 			error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
976 			    tcnp->cn_cred, tcnp->cn_thread);
977 		}
978 		xp->i_flag |= IN_CHANGE;
979 		vput(tvp);
980 		xp = NULL;
981 	}
982 
983 	/*
984 	 * 3) Unlink the source.
985 	 */
986 	fcnp->cn_flags &= ~MODMASK;
987 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
988 	VREF(fdvp);
989 	error = relookup(fdvp, &fvp, fcnp);
990 	if (error == 0)
991 		vrele(fdvp);
992 	if (fvp != NULL) {
993 		xp = VTOI(fvp);
994 		dp = VTOI(fdvp);
995 	} else {
996 		/*
997 		 * From name has disappeared.  IN_RENAME is not sufficient
998 		 * to protect against directory races due to timing windows,
999 		 * so we can't panic here.
1000 		 */
1001 		vrele(ap->a_fvp);
1002 		return (0);
1003 	}
1004 	/*
1005 	 * Ensure that the directory entry still exists and has not
1006 	 * changed while the new name has been entered. If the source is
1007 	 * a file then the entry may have been unlinked or renamed. In
1008 	 * either case there is no further work to be done. If the source
1009 	 * is a directory then it cannot have been rmdir'ed; its link
1010 	 * count of three would cause a rmdir to fail with ENOTEMPTY.
1011 	 * The IN_RENAME flag ensures that it cannot be moved by another
1012 	 * rename.
1013 	 */
1014 	if (xp != ip) {
1015 		/*
1016 		 * From name resolves to a different inode.  IN_RENAME is
1017 		 * not sufficient protection against timing window races
1018 		 * so we can't panic here.
1019 		 */
1020 	} else {
1021 		/*
1022 		 * If the source is a directory with a
1023 		 * new parent, the link count of the old
1024 		 * parent directory must be decremented
1025 		 * and ".." set to point to the new parent.
1026 		 */
1027 		if (doingdirectory && newparent) {
1028 			dp->i_nlink--;
1029 			dp->i_flag |= IN_CHANGE;
1030 			error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1031 			    sizeof(struct dirtemplate), (off_t)0,
1032 			    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1033 			    tcnp->cn_cred, NOCRED, NULL, NULL);
1034 			if (error == 0) {
1035 				/* Like ufs little-endian: */
1036 				namlen = dirbuf.dotdot_type;
1037 				if (namlen != 2 ||
1038 				    dirbuf.dotdot_name[0] != '.' ||
1039 				    dirbuf.dotdot_name[1] != '.') {
1040 					ext2_dirbad(xp, (doff_t)12,
1041 					    "rename: mangled dir");
1042 				} else {
1043 					dirbuf.dotdot_ino = newparent;
1044 					(void)vn_rdwr(UIO_WRITE, fvp,
1045 					    (caddr_t)&dirbuf,
1046 					    sizeof(struct dirtemplate),
1047 					    (off_t)0, UIO_SYSSPACE,
1048 					    IO_NODELOCKED | IO_SYNC |
1049 					    IO_NOMACCHECK, tcnp->cn_cred,
1050 					    NOCRED, NULL, NULL);
1051 					cache_purge(fdvp);
1052 				}
1053 			}
1054 		}
1055 		error = ext2_dirremove(fdvp, fcnp);
1056 		if (!error) {
1057 			xp->i_nlink--;
1058 			xp->i_flag |= IN_CHANGE;
1059 		}
1060 		xp->i_flag &= ~IN_RENAME;
1061 	}
1062 	if (dp)
1063 		vput(fdvp);
1064 	if (xp)
1065 		vput(fvp);
1066 	vrele(ap->a_fvp);
1067 	return (error);
1068 
1069 bad:
1070 	if (xp)
1071 		vput(ITOV(xp));
1072 	vput(ITOV(dp));
1073 out:
1074 	if (doingdirectory)
1075 		ip->i_flag &= ~IN_RENAME;
1076 	if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1077 		ip->i_nlink--;
1078 		ip->i_flag |= IN_CHANGE;
1079 		ip->i_flag &= ~IN_RENAME;
1080 		vput(fvp);
1081 	} else
1082 		vrele(fvp);
1083 	return (error);
1084 }
1085 
1086 /*
1087  * Mkdir system call
1088  */
1089 static int
1090 ext2_mkdir(struct vop_mkdir_args *ap)
1091 {
1092 	struct vnode *dvp = ap->a_dvp;
1093 	struct vattr *vap = ap->a_vap;
1094 	struct componentname *cnp = ap->a_cnp;
1095 	struct inode *ip, *dp;
1096 	struct vnode *tvp;
1097 	struct dirtemplate dirtemplate, *dtp;
1098 	int error, dmode;
1099 
1100 #ifdef INVARIANTS
1101 	if ((cnp->cn_flags & HASBUF) == 0)
1102 		panic("ext2_mkdir: no name");
1103 #endif
1104 	dp = VTOI(dvp);
1105 	if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
1106 		error = EMLINK;
1107 		goto out;
1108 	}
1109 	dmode = vap->va_mode & 0777;
1110 	dmode |= IFDIR;
1111 	/*
1112 	 * Must simulate part of ext2_makeinode here to acquire the inode,
1113 	 * but not have it entered in the parent directory. The entry is
1114 	 * made later after writing "." and ".." entries.
1115 	 */
1116 	error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1117 	if (error)
1118 		goto out;
1119 	ip = VTOI(tvp);
1120 	ip->i_gid = dp->i_gid;
1121 #ifdef SUIDDIR
1122 	{
1123 		/*
1124 		 * if we are hacking owners here, (only do this where told to)
1125 		 * and we are not giving it TOO root, (would subvert quotas)
1126 		 * then go ahead and give it to the other user.
1127 		 * The new directory also inherits the SUID bit.
1128 		 * If user's UID and dir UID are the same,
1129 		 * 'give it away' so that the SUID is still forced on.
1130 		 */
1131 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1132 		    (dp->i_mode & ISUID) && dp->i_uid) {
1133 			dmode |= ISUID;
1134 			ip->i_uid = dp->i_uid;
1135 		} else {
1136 			ip->i_uid = cnp->cn_cred->cr_uid;
1137 		}
1138 	}
1139 #else
1140 	ip->i_uid = cnp->cn_cred->cr_uid;
1141 #endif
1142 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1143 	ip->i_mode = dmode;
1144 	tvp->v_type = VDIR;	/* Rest init'd in getnewvnode(). */
1145 	ip->i_nlink = 2;
1146 	if (cnp->cn_flags & ISWHITEOUT)
1147 		ip->i_flags |= UF_OPAQUE;
1148 	error = ext2_update(tvp, 1);
1149 
1150 	/*
1151 	 * Bump link count in parent directory
1152 	 * to reflect work done below.  Should
1153 	 * be done before reference is created
1154 	 * so reparation is possible if we crash.
1155 	 */
1156 	dp->i_nlink++;
1157 	dp->i_flag |= IN_CHANGE;
1158 	error = ext2_update(dvp, !DOINGASYNC(dvp));
1159 	if (error)
1160 		goto bad;
1161 
1162 	/* Initialize directory with "." and ".." from static template. */
1163 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1164 	    EXT2F_INCOMPAT_FTYPE))
1165 		dtp = &mastertemplate;
1166 	else
1167 		dtp = &omastertemplate;
1168 	dirtemplate = *dtp;
1169 	dirtemplate.dot_ino = ip->i_number;
1170 	dirtemplate.dotdot_ino = dp->i_number;
1171 	/*
1172 	 * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's
1173 	 * just redefine it - for this function only
1174 	 */
1175 #undef  DIRBLKSIZ
1176 #define DIRBLKSIZ  VTOI(dvp)->i_e2fs->e2fs_bsize
1177 	dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1178 	error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1179 	    sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE,
1180 	    IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1181 	    NULL, NULL);
1182 	if (error) {
1183 		dp->i_nlink--;
1184 		dp->i_flag |= IN_CHANGE;
1185 		goto bad;
1186 	}
1187 	if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1188 		/* XXX should grow with balloc() */
1189 		panic("ext2_mkdir: blksize");
1190 	else {
1191 		ip->i_size = DIRBLKSIZ;
1192 		ip->i_flag |= IN_CHANGE;
1193 	}
1194 
1195 	/* Directory set up, now install its entry in the parent directory. */
1196 	error = ext2_direnter(ip, dvp, cnp);
1197 	if (error) {
1198 		dp->i_nlink--;
1199 		dp->i_flag |= IN_CHANGE;
1200 	}
1201 bad:
1202 	/*
1203 	 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1204 	 * for us because we set the link count to 0.
1205 	 */
1206 	if (error) {
1207 		ip->i_nlink = 0;
1208 		ip->i_flag |= IN_CHANGE;
1209 		vput(tvp);
1210 	} else
1211 		*ap->a_vpp = tvp;
1212 out:
1213 	return (error);
1214 #undef  DIRBLKSIZ
1215 #define DIRBLKSIZ  DEV_BSIZE
1216 }
1217 
1218 /*
1219  * Rmdir system call.
1220  */
1221 static int
1222 ext2_rmdir(struct vop_rmdir_args *ap)
1223 {
1224 	struct vnode *vp = ap->a_vp;
1225 	struct vnode *dvp = ap->a_dvp;
1226 	struct componentname *cnp = ap->a_cnp;
1227 	struct inode *ip, *dp;
1228 	int error;
1229 
1230 	ip = VTOI(vp);
1231 	dp = VTOI(dvp);
1232 
1233 	/*
1234 	 * Verify the directory is empty (and valid).
1235 	 * (Rmdir ".." won't be valid since
1236 	 *  ".." will contain a reference to
1237 	 *  the current directory and thus be
1238 	 *  non-empty.)
1239 	 */
1240 	if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1241 		error = ENOTEMPTY;
1242 		goto out;
1243 	}
1244 	if ((dp->i_flags & APPEND)
1245 	    || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1246 		error = EPERM;
1247 		goto out;
1248 	}
1249 	/*
1250 	 * Delete reference to directory before purging
1251 	 * inode.  If we crash in between, the directory
1252 	 * will be reattached to lost+found,
1253 	 */
1254 	error = ext2_dirremove(dvp, cnp);
1255 	if (error)
1256 		goto out;
1257 	dp->i_nlink--;
1258 	dp->i_flag |= IN_CHANGE;
1259 	cache_purge(dvp);
1260 	VOP_UNLOCK(dvp, 0);
1261 	/*
1262 	 * Truncate inode.  The only stuff left
1263 	 * in the directory is "." and "..".  The
1264 	 * "." reference is inconsequential since
1265 	 * we're quashing it.  The ".." reference
1266 	 * has already been adjusted above.  We've
1267 	 * removed the "." reference and the reference
1268 	 * in the parent directory, but there may be
1269 	 * other hard links so decrement by 2 and
1270 	 * worry about them later.
1271 	 */
1272 	ip->i_nlink -= 2;
1273 	error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1274 	    cnp->cn_thread);
1275 	cache_purge(ITOV(ip));
1276 	if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1277 		VOP_UNLOCK(vp, 0);
1278 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1279 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1280 	}
1281 out:
1282 	return (error);
1283 }
1284 
1285 /*
1286  * symlink -- make a symbolic link
1287  */
1288 static int
1289 ext2_symlink(struct vop_symlink_args *ap)
1290 {
1291 	struct vnode *vp, **vpp = ap->a_vpp;
1292 	struct inode *ip;
1293 	int len, error;
1294 
1295 	error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1296 	    vpp, ap->a_cnp);
1297 	if (error)
1298 		return (error);
1299 	vp = *vpp;
1300 	len = strlen(ap->a_target);
1301 	if (len < vp->v_mount->mnt_maxsymlinklen) {
1302 		ip = VTOI(vp);
1303 		bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1304 		ip->i_size = len;
1305 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1306 	} else
1307 		error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1308 		    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1309 		    ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1310 	if (error)
1311 		vput(vp);
1312 	return (error);
1313 }
1314 
1315 /*
1316  * Return target name of a symbolic link
1317  */
1318 static int
1319 ext2_readlink(struct vop_readlink_args *ap)
1320 {
1321 	struct vnode *vp = ap->a_vp;
1322 	struct inode *ip = VTOI(vp);
1323 	int isize;
1324 
1325 	isize = ip->i_size;
1326 	if (isize < vp->v_mount->mnt_maxsymlinklen) {
1327 		uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1328 		return (0);
1329 	}
1330 	return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1331 }
1332 
1333 /*
1334  * Calculate the logical to physical mapping if not done already,
1335  * then call the device strategy routine.
1336  *
1337  * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1338  * deadlock on memory.  See ext2_bmap() for details.
1339  */
1340 static int
1341 ext2_strategy(struct vop_strategy_args *ap)
1342 {
1343 	struct buf *bp = ap->a_bp;
1344 	struct vnode *vp = ap->a_vp;
1345 	struct bufobj *bo;
1346 	daddr_t blkno;
1347 	int error;
1348 
1349 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1350 		panic("ext2_strategy: spec");
1351 	if (bp->b_blkno == bp->b_lblkno) {
1352 		error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1353 		bp->b_blkno = blkno;
1354 		if (error) {
1355 			bp->b_error = error;
1356 			bp->b_ioflags |= BIO_ERROR;
1357 			bufdone(bp);
1358 			return (0);
1359 		}
1360 		if ((long)bp->b_blkno == -1)
1361 			vfs_bio_clrbuf(bp);
1362 	}
1363 	if ((long)bp->b_blkno == -1) {
1364 		bufdone(bp);
1365 		return (0);
1366 	}
1367 	bp->b_iooffset = dbtob(bp->b_blkno);
1368 	bo = VFSTOEXT2(vp->v_mount)->um_bo;
1369 	BO_STRATEGY(bo, bp);
1370 	return (0);
1371 }
1372 
1373 /*
1374  * Print out the contents of an inode.
1375  */
1376 static int
1377 ext2_print(struct vop_print_args *ap)
1378 {
1379 	struct vnode *vp = ap->a_vp;
1380 	struct inode *ip = VTOI(vp);
1381 
1382 	vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number);
1383 	if (vp->v_type == VFIFO)
1384 		fifo_printinfo(vp);
1385 	printf("\n");
1386 	return (0);
1387 }
1388 
1389 /*
1390  * Close wrapper for fifos.
1391  *
1392  * Update the times on the inode then do device close.
1393  */
1394 static int
1395 ext2fifo_close(struct vop_close_args *ap)
1396 {
1397 	struct vnode *vp = ap->a_vp;
1398 
1399 	VI_LOCK(vp);
1400 	if (vp->v_usecount > 1)
1401 		ext2_itimes_locked(vp);
1402 	VI_UNLOCK(vp);
1403 	return (fifo_specops.vop_close(ap));
1404 }
1405 
1406 /*
1407  * Kqfilter wrapper for fifos.
1408  *
1409  * Fall through to ext2 kqfilter routines if needed
1410  */
1411 static int
1412 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1413 {
1414 	int error;
1415 
1416 	error = fifo_specops.vop_kqfilter(ap);
1417 	if (error)
1418 		error = vfs_kqfilter(ap);
1419 	return (error);
1420 }
1421 
1422 /*
1423  * Return POSIX pathconf information applicable to ext2 filesystems.
1424  */
1425 static int
1426 ext2_pathconf(struct vop_pathconf_args *ap)
1427 {
1428 	int error = 0;
1429 
1430 	switch (ap->a_name) {
1431 	case _PC_LINK_MAX:
1432 		*ap->a_retval = EXT2_LINK_MAX;
1433 		break;
1434 	case _PC_NAME_MAX:
1435 		*ap->a_retval = NAME_MAX;
1436 		break;
1437 	case _PC_PATH_MAX:
1438 		*ap->a_retval = PATH_MAX;
1439 		break;
1440 	case _PC_PIPE_BUF:
1441 		*ap->a_retval = PIPE_BUF;
1442 		break;
1443 	case _PC_CHOWN_RESTRICTED:
1444 		*ap->a_retval = 1;
1445 		break;
1446 	case _PC_NO_TRUNC:
1447 		*ap->a_retval = 1;
1448 		break;
1449 	case _PC_MIN_HOLE_SIZE:
1450 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1451 		break;
1452 	case _PC_ASYNC_IO:
1453 		/* _PC_ASYNC_IO should have been handled by upper layers. */
1454 		KASSERT(0, ("_PC_ASYNC_IO should not get here"));
1455 		error = EINVAL;
1456 		break;
1457 	case _PC_PRIO_IO:
1458 		*ap->a_retval = 0;
1459 		break;
1460 	case _PC_SYNC_IO:
1461 		*ap->a_retval = 0;
1462 		break;
1463 	case _PC_ALLOC_SIZE_MIN:
1464 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1465 		break;
1466 	case _PC_FILESIZEBITS:
1467 		*ap->a_retval = 64;
1468 		break;
1469 	case _PC_REC_INCR_XFER_SIZE:
1470 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1471 		break;
1472 	case _PC_REC_MAX_XFER_SIZE:
1473 		*ap->a_retval = -1;	/* means ``unlimited'' */
1474 		break;
1475 	case _PC_REC_MIN_XFER_SIZE:
1476 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1477 		break;
1478 	case _PC_REC_XFER_ALIGN:
1479 		*ap->a_retval = PAGE_SIZE;
1480 		break;
1481 	case _PC_SYMLINK_MAX:
1482 		*ap->a_retval = MAXPATHLEN;
1483 		break;
1484 
1485 	default:
1486 		error = EINVAL;
1487 		break;
1488 	}
1489 	return (error);
1490 }
1491 
1492 /*
1493  * Vnode operation to remove a named attribute.
1494  */
1495 static int
1496 ext2_deleteextattr(struct vop_deleteextattr_args *ap)
1497 {
1498 	struct inode *ip;
1499 	struct m_ext2fs *fs;
1500 	int error;
1501 
1502 	ip = VTOI(ap->a_vp);
1503 	fs = ip->i_e2fs;
1504 
1505 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1506 		return (EOPNOTSUPP);
1507 
1508 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1509 		return (EOPNOTSUPP);
1510 
1511 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1512 	    ap->a_cred, ap->a_td, VWRITE);
1513 	if (error)
1514 		return (error);
1515 
1516 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1517 		error = ext2_extattr_inode_delete(ip, ap->a_attrnamespace, ap->a_name);
1518 		if (error != ENOATTR)
1519 			return (error);
1520 	}
1521 
1522 	if (ip->i_facl)
1523 		error = ext2_extattr_block_delete(ip, ap->a_attrnamespace, ap->a_name);
1524 
1525 	return (error);
1526 }
1527 
1528 /*
1529  * Vnode operation to retrieve a named extended attribute.
1530  */
1531 static int
1532 ext2_getextattr(struct vop_getextattr_args *ap)
1533 {
1534 	struct inode *ip;
1535 	struct m_ext2fs *fs;
1536 	int error;
1537 
1538 	ip = VTOI(ap->a_vp);
1539 	fs = ip->i_e2fs;
1540 
1541 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1542 		return (EOPNOTSUPP);
1543 
1544 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1545 		return (EOPNOTSUPP);
1546 
1547 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1548 	    ap->a_cred, ap->a_td, VREAD);
1549 	if (error)
1550 		return (error);
1551 
1552 	if (ap->a_size != NULL)
1553 		*ap->a_size = 0;
1554 
1555 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1556 		error = ext2_extattr_inode_get(ip, ap->a_attrnamespace,
1557 		    ap->a_name, ap->a_uio, ap->a_size);
1558 		if (error != ENOATTR)
1559 			return (error);
1560 	}
1561 
1562 	if (ip->i_facl)
1563 		error = ext2_extattr_block_get(ip, ap->a_attrnamespace,
1564 		    ap->a_name, ap->a_uio, ap->a_size);
1565 
1566 	return (error);
1567 }
1568 
1569 /*
1570  * Vnode operation to retrieve extended attributes on a vnode.
1571  */
1572 static int
1573 ext2_listextattr(struct vop_listextattr_args *ap)
1574 {
1575 	struct inode *ip;
1576 	struct m_ext2fs *fs;
1577 	int error;
1578 
1579 	ip = VTOI(ap->a_vp);
1580 	fs = ip->i_e2fs;
1581 
1582 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1583 		return (EOPNOTSUPP);
1584 
1585 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1586 		return (EOPNOTSUPP);
1587 
1588 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1589 	    ap->a_cred, ap->a_td, VREAD);
1590 	if (error)
1591 		return (error);
1592 
1593 	if (ap->a_size != NULL)
1594 		*ap->a_size = 0;
1595 
1596 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1597 		error = ext2_extattr_inode_list(ip, ap->a_attrnamespace,
1598 		    ap->a_uio, ap->a_size);
1599 		if (error)
1600 			return (error);
1601 	}
1602 
1603 	if (ip->i_facl)
1604 		error = ext2_extattr_block_list(ip, ap->a_attrnamespace,
1605 		    ap->a_uio, ap->a_size);
1606 
1607 	return (error);
1608 }
1609 
1610 /*
1611  * Vnode operation to set a named attribute.
1612  */
1613 static int
1614 ext2_setextattr(struct vop_setextattr_args *ap)
1615 {
1616 	struct inode *ip;
1617 	struct m_ext2fs *fs;
1618 	int error;
1619 
1620 	ip = VTOI(ap->a_vp);
1621 	fs = ip->i_e2fs;
1622 
1623 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1624 		return (EOPNOTSUPP);
1625 
1626 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1627 		return (EOPNOTSUPP);
1628 
1629 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1630 	    ap->a_cred, ap->a_td, VWRITE);
1631 	if (error)
1632 		return (error);
1633 
1634 	error = ext2_extattr_valid_attrname(ap->a_attrnamespace, ap->a_name);
1635 	if (error)
1636 		return (error);
1637 
1638 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1639 		error = ext2_extattr_inode_set(ip, ap->a_attrnamespace,
1640 		    ap->a_name, ap->a_uio);
1641 		if (error != ENOSPC)
1642 			return (error);
1643 	}
1644 
1645 	error = ext2_extattr_block_set(ip, ap->a_attrnamespace,
1646 	    ap->a_name, ap->a_uio);
1647 
1648 	return (error);
1649 }
1650 
1651 /*
1652  * Vnode pointer to File handle
1653  */
1654 /* ARGSUSED */
1655 static int
1656 ext2_vptofh(struct vop_vptofh_args *ap)
1657 {
1658 	struct inode *ip;
1659 	struct ufid *ufhp;
1660 
1661 	ip = VTOI(ap->a_vp);
1662 	ufhp = (struct ufid *)ap->a_fhp;
1663 	ufhp->ufid_len = sizeof(struct ufid);
1664 	ufhp->ufid_ino = ip->i_number;
1665 	ufhp->ufid_gen = ip->i_gen;
1666 	return (0);
1667 }
1668 
1669 /*
1670  * Initialize the vnode associated with a new inode, handle aliased
1671  * vnodes.
1672  */
1673 int
1674 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1675 {
1676 	struct inode *ip;
1677 	struct vnode *vp;
1678 
1679 	vp = *vpp;
1680 	ip = VTOI(vp);
1681 	vp->v_type = IFTOVT(ip->i_mode);
1682 	if (vp->v_type == VFIFO)
1683 		vp->v_op = fifoops;
1684 
1685 	if (ip->i_number == EXT2_ROOTINO)
1686 		vp->v_vflag |= VV_ROOT;
1687 	ip->i_modrev = init_va_filerev();
1688 	*vpp = vp;
1689 	return (0);
1690 }
1691 
1692 /*
1693  * Allocate a new inode.
1694  */
1695 static int
1696 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1697     struct componentname *cnp)
1698 {
1699 	struct inode *ip, *pdir;
1700 	struct vnode *tvp;
1701 	int error;
1702 
1703 	pdir = VTOI(dvp);
1704 #ifdef INVARIANTS
1705 	if ((cnp->cn_flags & HASBUF) == 0)
1706 		panic("ext2_makeinode: no name");
1707 #endif
1708 	*vpp = NULL;
1709 	if ((mode & IFMT) == 0)
1710 		mode |= IFREG;
1711 
1712 	error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1713 	if (error) {
1714 		return (error);
1715 	}
1716 	ip = VTOI(tvp);
1717 	ip->i_gid = pdir->i_gid;
1718 #ifdef SUIDDIR
1719 	{
1720 		/*
1721 		 * if we are
1722 		 * not the owner of the directory,
1723 		 * and we are hacking owners here, (only do this where told to)
1724 		 * and we are not giving it TOO root, (would subvert quotas)
1725 		 * then go ahead and give it to the other user.
1726 		 * Note that this drops off the execute bits for security.
1727 		 */
1728 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1729 		    (pdir->i_mode & ISUID) &&
1730 		    (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1731 			ip->i_uid = pdir->i_uid;
1732 			mode &= ~07111;
1733 		} else {
1734 			ip->i_uid = cnp->cn_cred->cr_uid;
1735 		}
1736 	}
1737 #else
1738 	ip->i_uid = cnp->cn_cred->cr_uid;
1739 #endif
1740 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1741 	ip->i_mode = mode;
1742 	tvp->v_type = IFTOVT(mode);	/* Rest init'd in getnewvnode(). */
1743 	ip->i_nlink = 1;
1744 	if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1745 		if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1746 			ip->i_mode &= ~ISGID;
1747 	}
1748 
1749 	if (cnp->cn_flags & ISWHITEOUT)
1750 		ip->i_flags |= UF_OPAQUE;
1751 
1752 	/*
1753 	 * Make sure inode goes to disk before directory entry.
1754 	 */
1755 	error = ext2_update(tvp, !DOINGASYNC(tvp));
1756 	if (error)
1757 		goto bad;
1758 	error = ext2_direnter(ip, dvp, cnp);
1759 	if (error)
1760 		goto bad;
1761 
1762 	*vpp = tvp;
1763 	return (0);
1764 
1765 bad:
1766 	/*
1767 	 * Write error occurred trying to update the inode
1768 	 * or the directory so must deallocate the inode.
1769 	 */
1770 	ip->i_nlink = 0;
1771 	ip->i_flag |= IN_CHANGE;
1772 	vput(tvp);
1773 	return (error);
1774 }
1775 
1776 /*
1777  * Vnode op for reading.
1778  */
1779 static int
1780 ext2_read(struct vop_read_args *ap)
1781 {
1782 	struct vnode *vp;
1783 	struct inode *ip;
1784 	int error;
1785 
1786 	vp = ap->a_vp;
1787 	ip = VTOI(vp);
1788 
1789 	/* EXT4_EXT_LOCK(ip); */
1790 	if (ip->i_flag & IN_E4EXTENTS)
1791 		error = ext4_ext_read(ap);
1792 	else
1793 		error = ext2_ind_read(ap);
1794 	/* EXT4_EXT_UNLOCK(ip); */
1795 	return (error);
1796 }
1797 
1798 /*
1799  * Vnode op for reading.
1800  */
1801 static int
1802 ext2_ind_read(struct vop_read_args *ap)
1803 {
1804 	struct vnode *vp;
1805 	struct inode *ip;
1806 	struct uio *uio;
1807 	struct m_ext2fs *fs;
1808 	struct buf *bp;
1809 	daddr_t lbn, nextlbn;
1810 	off_t bytesinfile;
1811 	long size, xfersize, blkoffset;
1812 	int error, orig_resid, seqcount;
1813 	int ioflag;
1814 
1815 	vp = ap->a_vp;
1816 	uio = ap->a_uio;
1817 	ioflag = ap->a_ioflag;
1818 
1819 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
1820 	ip = VTOI(vp);
1821 
1822 #ifdef INVARIANTS
1823 	if (uio->uio_rw != UIO_READ)
1824 		panic("%s: mode", "ext2_read");
1825 
1826 	if (vp->v_type == VLNK) {
1827 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
1828 			panic("%s: short symlink", "ext2_read");
1829 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
1830 		panic("%s: type %d", "ext2_read", vp->v_type);
1831 #endif
1832 	orig_resid = uio->uio_resid;
1833 	KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
1834 	if (orig_resid == 0)
1835 		return (0);
1836 	KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
1837 	fs = ip->i_e2fs;
1838 	if (uio->uio_offset < ip->i_size &&
1839 	    uio->uio_offset >= fs->e2fs_maxfilesize)
1840 		return (EOVERFLOW);
1841 
1842 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1843 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1844 			break;
1845 		lbn = lblkno(fs, uio->uio_offset);
1846 		nextlbn = lbn + 1;
1847 		size = blksize(fs, ip, lbn);
1848 		blkoffset = blkoff(fs, uio->uio_offset);
1849 
1850 		xfersize = fs->e2fs_fsize - blkoffset;
1851 		if (uio->uio_resid < xfersize)
1852 			xfersize = uio->uio_resid;
1853 		if (bytesinfile < xfersize)
1854 			xfersize = bytesinfile;
1855 
1856 		if (lblktosize(fs, nextlbn) >= ip->i_size)
1857 			error = bread(vp, lbn, size, NOCRED, &bp);
1858 		else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
1859 			error = cluster_read(vp, ip->i_size, lbn, size,
1860 			    NOCRED, blkoffset + uio->uio_resid, seqcount,
1861 			    0, &bp);
1862 		} else if (seqcount > 1) {
1863 			u_int nextsize = blksize(fs, ip, nextlbn);
1864 
1865 			error = breadn(vp, lbn,
1866 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1867 		} else
1868 			error = bread(vp, lbn, size, NOCRED, &bp);
1869 		if (error) {
1870 			brelse(bp);
1871 			bp = NULL;
1872 			break;
1873 		}
1874 
1875 		/*
1876 		 * We should only get non-zero b_resid when an I/O error
1877 		 * has occurred, which should cause us to break above.
1878 		 * However, if the short read did not cause an error,
1879 		 * then we want to ensure that we do not uiomove bad
1880 		 * or uninitialized data.
1881 		 */
1882 		size -= bp->b_resid;
1883 		if (size < xfersize) {
1884 			if (size == 0)
1885 				break;
1886 			xfersize = size;
1887 		}
1888 		error = uiomove((char *)bp->b_data + blkoffset,
1889 		    (int)xfersize, uio);
1890 		if (error)
1891 			break;
1892 		vfs_bio_brelse(bp, ioflag);
1893 	}
1894 
1895 	/*
1896 	 * This can only happen in the case of an error because the loop
1897 	 * above resets bp to NULL on each iteration and on normal
1898 	 * completion has not set a new value into it. so it must have come
1899 	 * from a 'break' statement
1900 	 */
1901 	if (bp != NULL)
1902 		vfs_bio_brelse(bp, ioflag);
1903 
1904 	if ((error == 0 || uio->uio_resid != orig_resid) &&
1905 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
1906 		ip->i_flag |= IN_ACCESS;
1907 	return (error);
1908 }
1909 
1910 static int
1911 ext2_ioctl(struct vop_ioctl_args *ap)
1912 {
1913 
1914 	switch (ap->a_command) {
1915 	case FIOSEEKDATA:
1916 	case FIOSEEKHOLE:
1917 		return (vn_bmap_seekhole(ap->a_vp, ap->a_command,
1918 		    (off_t *)ap->a_data, ap->a_cred));
1919 	default:
1920 		return (ENOTTY);
1921 	}
1922 }
1923 
1924 /*
1925  * this function handles ext4 extents block mapping
1926  */
1927 static int
1928 ext4_ext_read(struct vop_read_args *ap)
1929 {
1930 	static unsigned char zeroes[EXT2_MAX_BLOCK_SIZE];
1931 	struct vnode *vp;
1932 	struct inode *ip;
1933 	struct uio *uio;
1934 	struct m_ext2fs *fs;
1935 	struct buf *bp;
1936 	struct ext4_extent nex, *ep;
1937 	struct ext4_extent_path path;
1938 	daddr_t lbn, newblk;
1939 	off_t bytesinfile;
1940 	int cache_type;
1941 	ssize_t orig_resid;
1942 	int error;
1943 	long size, xfersize, blkoffset;
1944 
1945 	vp = ap->a_vp;
1946 	ip = VTOI(vp);
1947 	uio = ap->a_uio;
1948 	memset(&path, 0, sizeof(path));
1949 
1950 	orig_resid = uio->uio_resid;
1951 	KASSERT(orig_resid >= 0, ("%s: uio->uio_resid < 0", __func__));
1952 	if (orig_resid == 0)
1953 		return (0);
1954 	KASSERT(uio->uio_offset >= 0, ("%s: uio->uio_offset < 0", __func__));
1955 	fs = ip->i_e2fs;
1956 	if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize)
1957 		return (EOVERFLOW);
1958 
1959 	while (uio->uio_resid > 0) {
1960 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1961 			break;
1962 		lbn = lblkno(fs, uio->uio_offset);
1963 		size = blksize(fs, ip, lbn);
1964 		blkoffset = blkoff(fs, uio->uio_offset);
1965 
1966 		xfersize = fs->e2fs_fsize - blkoffset;
1967 		xfersize = MIN(xfersize, uio->uio_resid);
1968 		xfersize = MIN(xfersize, bytesinfile);
1969 
1970 		/* get block from ext4 extent cache */
1971 		cache_type = ext4_ext_in_cache(ip, lbn, &nex);
1972 		switch (cache_type) {
1973 		case EXT4_EXT_CACHE_NO:
1974 			ext4_ext_find_extent(fs, ip, lbn, &path);
1975 			if (path.ep_is_sparse)
1976 				ep = &path.ep_sparse_ext;
1977 			else
1978 				ep = path.ep_ext;
1979 			if (ep == NULL)
1980 				return (EIO);
1981 
1982 			ext4_ext_put_cache(ip, ep,
1983 			    path.ep_is_sparse ? EXT4_EXT_CACHE_GAP : EXT4_EXT_CACHE_IN);
1984 
1985 			newblk = lbn - ep->e_blk + (ep->e_start_lo |
1986 			    (daddr_t)ep->e_start_hi << 32);
1987 
1988 			if (path.ep_bp != NULL) {
1989 				brelse(path.ep_bp);
1990 				path.ep_bp = NULL;
1991 			}
1992 			break;
1993 
1994 		case EXT4_EXT_CACHE_GAP:
1995 			/* block has not been allocated yet */
1996 			break;
1997 
1998 		case EXT4_EXT_CACHE_IN:
1999 			newblk = lbn - nex.e_blk + (nex.e_start_lo |
2000 			    (daddr_t)nex.e_start_hi << 32);
2001 			break;
2002 
2003 		default:
2004 			panic("%s: invalid cache type", __func__);
2005 		}
2006 
2007 		if (cache_type == EXT4_EXT_CACHE_GAP ||
2008 		    (cache_type == EXT4_EXT_CACHE_NO && path.ep_is_sparse)) {
2009 			if (xfersize > sizeof(zeroes))
2010 				xfersize = sizeof(zeroes);
2011 			error = uiomove(zeroes, xfersize, uio);
2012 			if (error)
2013 				return (error);
2014 		} else {
2015 			error = bread(ip->i_devvp, fsbtodb(fs, newblk), size,
2016 			    NOCRED, &bp);
2017 			if (error) {
2018 				brelse(bp);
2019 				return (error);
2020 			}
2021 
2022 			size -= bp->b_resid;
2023 			if (size < xfersize) {
2024 				if (size == 0) {
2025 					bqrelse(bp);
2026 					break;
2027 				}
2028 				xfersize = size;
2029 			}
2030 			error = uiomove(bp->b_data + blkoffset, xfersize, uio);
2031 			bqrelse(bp);
2032 			if (error)
2033 				return (error);
2034 		}
2035 	}
2036 
2037 	return (0);
2038 }
2039 
2040 /*
2041  * Vnode op for writing.
2042  */
2043 static int
2044 ext2_write(struct vop_write_args *ap)
2045 {
2046 	struct vnode *vp;
2047 	struct uio *uio;
2048 	struct inode *ip;
2049 	struct m_ext2fs *fs;
2050 	struct buf *bp;
2051 	daddr_t lbn;
2052 	off_t osize;
2053 	int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
2054 
2055 	ioflag = ap->a_ioflag;
2056 	uio = ap->a_uio;
2057 	vp = ap->a_vp;
2058 
2059 	seqcount = ioflag >> IO_SEQSHIFT;
2060 	ip = VTOI(vp);
2061 
2062 #ifdef INVARIANTS
2063 	if (uio->uio_rw != UIO_WRITE)
2064 		panic("%s: mode", "ext2_write");
2065 #endif
2066 
2067 	switch (vp->v_type) {
2068 	case VREG:
2069 		if (ioflag & IO_APPEND)
2070 			uio->uio_offset = ip->i_size;
2071 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
2072 			return (EPERM);
2073 		/* FALLTHROUGH */
2074 	case VLNK:
2075 		break;
2076 	case VDIR:
2077 		/* XXX differs from ffs -- this is called from ext2_mkdir(). */
2078 		if ((ioflag & IO_SYNC) == 0)
2079 			panic("ext2_write: nonsync dir write");
2080 		break;
2081 	default:
2082 		panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
2083 		    vp->v_type, (intmax_t)uio->uio_offset,
2084 		    (intmax_t)uio->uio_resid);
2085 	}
2086 
2087 	KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
2088 	KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
2089 	fs = ip->i_e2fs;
2090 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
2091 		return (EFBIG);
2092 	/*
2093 	 * Maybe this should be above the vnode op call, but so long as
2094 	 * file servers have no limits, I don't think it matters.
2095 	 */
2096 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
2097 		return (EFBIG);
2098 
2099 	resid = uio->uio_resid;
2100 	osize = ip->i_size;
2101 	if (seqcount > BA_SEQMAX)
2102 		flags = BA_SEQMAX << BA_SEQSHIFT;
2103 	else
2104 		flags = seqcount << BA_SEQSHIFT;
2105 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
2106 		flags |= IO_SYNC;
2107 
2108 	for (error = 0; uio->uio_resid > 0;) {
2109 		lbn = lblkno(fs, uio->uio_offset);
2110 		blkoffset = blkoff(fs, uio->uio_offset);
2111 		xfersize = fs->e2fs_fsize - blkoffset;
2112 		if (uio->uio_resid < xfersize)
2113 			xfersize = uio->uio_resid;
2114 		if (uio->uio_offset + xfersize > ip->i_size)
2115 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
2116 
2117 		/*
2118 		 * We must perform a read-before-write if the transfer size
2119 		 * does not cover the entire buffer.
2120 		 */
2121 		if (fs->e2fs_bsize > xfersize)
2122 			flags |= BA_CLRBUF;
2123 		else
2124 			flags &= ~BA_CLRBUF;
2125 		error = ext2_balloc(ip, lbn, blkoffset + xfersize,
2126 		    ap->a_cred, &bp, flags);
2127 		if (error != 0)
2128 			break;
2129 
2130 		if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL))
2131 			bp->b_flags |= B_NOCACHE;
2132 		if (uio->uio_offset + xfersize > ip->i_size)
2133 			ip->i_size = uio->uio_offset + xfersize;
2134 		size = blksize(fs, ip, lbn) - bp->b_resid;
2135 		if (size < xfersize)
2136 			xfersize = size;
2137 
2138 		error =
2139 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
2140 		/*
2141 		 * If the buffer is not already filled and we encounter an
2142 		 * error while trying to fill it, we have to clear out any
2143 		 * garbage data from the pages instantiated for the buffer.
2144 		 * If we do not, a failed uiomove() during a write can leave
2145 		 * the prior contents of the pages exposed to a userland mmap.
2146 		 *
2147 		 * Note that we need only clear buffers with a transfer size
2148 		 * equal to the block size because buffers with a shorter
2149 		 * transfer size were cleared above by the call to ext2_balloc()
2150 		 * with the BA_CLRBUF flag set.
2151 		 *
2152 		 * If the source region for uiomove identically mmaps the
2153 		 * buffer, uiomove() performed the NOP copy, and the buffer
2154 		 * content remains valid because the page fault handler
2155 		 * validated the pages.
2156 		 */
2157 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2158 		    fs->e2fs_bsize == xfersize)
2159 			vfs_bio_clrbuf(bp);
2160 
2161 		vfs_bio_set_flags(bp, ioflag);
2162 
2163 		/*
2164 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
2165 		 * if we have a severe page deficiency write the buffer
2166 		 * asynchronously.  Otherwise try to cluster, and if that
2167 		 * doesn't do it then either do an async write (if O_DIRECT),
2168 		 * or a delayed write (if not).
2169 		 */
2170 		if (ioflag & IO_SYNC) {
2171 			(void)bwrite(bp);
2172 		} else if (vm_page_count_severe() ||
2173 			    buf_dirty_count_severe() ||
2174 		    (ioflag & IO_ASYNC)) {
2175 			bp->b_flags |= B_CLUSTEROK;
2176 			bawrite(bp);
2177 		} else if (xfersize + blkoffset == fs->e2fs_fsize) {
2178 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2179 				bp->b_flags |= B_CLUSTEROK;
2180 				cluster_write(vp, bp, ip->i_size, seqcount, 0);
2181 			} else {
2182 				bawrite(bp);
2183 			}
2184 		} else if (ioflag & IO_DIRECT) {
2185 			bp->b_flags |= B_CLUSTEROK;
2186 			bawrite(bp);
2187 		} else {
2188 			bp->b_flags |= B_CLUSTEROK;
2189 			bdwrite(bp);
2190 		}
2191 		if (error || xfersize == 0)
2192 			break;
2193 	}
2194 	/*
2195 	 * If we successfully wrote any data, and we are not the superuser
2196 	 * we clear the setuid and setgid bits as a precaution against
2197 	 * tampering.
2198 	 */
2199 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2200 	    ap->a_cred) {
2201 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
2202 			ip->i_mode &= ~(ISUID | ISGID);
2203 	}
2204 	if (error) {
2205 		if (ioflag & IO_UNIT) {
2206 			(void)ext2_truncate(vp, osize,
2207 			    ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2208 			uio->uio_offset -= resid - uio->uio_resid;
2209 			uio->uio_resid = resid;
2210 		}
2211 	}
2212 	if (uio->uio_resid != resid) {
2213 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
2214 		if (ioflag & IO_SYNC)
2215 			error = ext2_update(vp, 1);
2216 	}
2217 	return (error);
2218 }
2219